Initial setup

In [1]:
#Import libraries
import pandas as pd
import numpy as np
import plotly.express as px
import seaborn as sns
import matplotlib.pyplot as plt
from sklearn.preprocessing import StandardScaler
import missingno as msno
print("Packages uploaded successfully!")

Packages uploaded successfully!
In [2]:
#Read dataset
data = pd.read_csv('./machine_readable_dnt.csv')
In [3]:
#Check first five rows
data.head(5)
Out[3]:
ID Chemical CASRN DNT SLogP BBB Cbrain/Cblood Pgp_inhibition Pgp_substrate Pgp_active BDNF, Reduction Activity_Syn Syn_EC30 Viability_EC30 Activity_NNF NNF EC50min NNF EC50max Viability_LDH
0 1 5-Fluorouracil 51-21-8 1 -0.7977 1.0 0.199528 0.0 0.0 0.0 1.0 1 0.0 91.8 1 0.46 5.03 3.37
1 2 Acrylamide 79-06-1 1 -0.3423 1.0 0.380063 0.0 0.0 0.0 1.0 0 0.0 0.0 1 8.26 11.30 0.00
2 3 Aldicarb 116-06-3 1 1.4698 1.0 0.609681 0.0 0.0 0.0 NaN 0 0.0 0.0 1 0.66 9.88 0.00
3 4 Aminonicotinamide 329-89-5 1 -0.2373 1.0 0.212859 0.0 0.0 0.0 NaN 1 7.9 6.7 1 3.18 9.85 6.27
4 5 Bisphenol A 80-05-7 1 3.4237 1.0 6.030440 0.0 0.0 0.0 1.0 1 33.9 95.0 1 1.01 25.17 0.00
In [4]:
#Statistical summary
data_stats = data.drop(["ID"], axis=1)
data_stats.describe()
Out[4]:
DNT SLogP BBB Cbrain/Cblood Pgp_inhibition Pgp_substrate Pgp_active BDNF, Reduction Activity_Syn Syn_EC30 Viability_EC30 Activity_NNF NNF EC50min NNF EC50max Viability_LDH
count 97.000000 95.000000 93.000000 87.000000 64.000000 64.000000 64.000000 37.000000 97.000000 67.000000 67.000000 97.000000 86.000000 86.000000 79.000000
mean 0.752577 1.942219 0.870968 2.179086 0.125000 0.078125 0.171875 0.594595 0.360825 6.077761 7.157015 0.659794 4.154419 33.273721 41.174810
std 0.433756 3.159165 0.337053 3.948392 0.333333 0.270490 0.380254 0.497743 0.482735 11.874245 17.468629 0.476240 20.399918 214.138809 332.030848
min 0.000000 -5.994500 0.000000 0.008951 0.000000 0.000000 0.000000 0.000000 0.000000 0.000000 0.000000 0.000000 0.000000 0.000000 0.000000
25% 1.000000 0.080800 1.000000 0.349445 0.000000 0.000000 0.000000 0.000000 0.000000 0.000000 0.000000 0.000000 0.002500 0.007500 0.000000
50% 1.000000 1.867700 1.000000 0.798734 0.000000 0.000000 0.000000 1.000000 0.000000 0.000000 0.000000 1.000000 0.105000 6.510000 0.000000
75% 1.000000 4.443850 1.000000 1.598055 0.000000 0.000000 0.000000 1.000000 1.000000 7.100000 5.850000 1.000000 1.307500 21.907500 2.125000
max 1.000000 9.694600 1.000000 23.155500 1.000000 1.000000 1.000000 1.000000 1.000000 51.900000 95.000000 1.000000 186.850000 1993.650000 2954.160000

Distribution of categorical variables

In [5]:
#DNT
replace_values = {0 : 'Negative/Safe', 1 : 'Positive'}
dnt = data.replace({"DNT": replace_values})
#sns.set_palette("Set3") #or pastel
#sns.set_palette("Reds")
my_palette = ["#e74c3c", "#2ecc71"]
sns.set_palette(sns.color_palette(my_palette))
ax = sns.countplot(x="DNT", data=dnt)
ax.set(ylabel='Number of compounds', xlabel=None, title ='DNT Classification')
sns.despine(right=True, bottom = True)
In [6]:
#BBB
replace_values_bbb = {0 : 'Unpermeable', 1 : 'Permeable'}
bbb = data.replace({"BBB": replace_values_bbb})
my_palette = ["#e74c3c", "#2ecc71"]
sns.set_palette(sns.color_palette(my_palette))
ax = sns.countplot(x="BBB", data=bbb)
ax.set(ylabel='Number of compounds', xlabel=None, title ='Blood-brain-barrier Permeability')
sns.despine(right=True, bottom = True)
In [7]:
#Pgp_inhibition
replace_values_pgp_inh = {1 : 'Inhibitor', 0 : 'Non-inhibitor'}
pgp_inh = data.replace({"Pgp_inhibition": replace_values_pgp_inh})
my_palette = ["#e74c3c", "#2ecc71" ]
sns.set_palette(sns.color_palette(my_palette))
ax = sns.countplot(x="Pgp_inhibition", data=pgp_inh, order = ["Inhibitor", "Non-inhibitor"])
ax.set(ylabel='Number of compounds', xlabel=None, title ='Inhibition of P-glycoprotein')
sns.despine(right=True, bottom = True)
In [8]:
#Pgp_substrate
replace_values_pgp_subs = {0 : 'Non-substrate', 1 : 'Substrate'}
pgp_subs = data.replace({"Pgp_substrate": replace_values_pgp_subs})
my_palette = ["#e74c3c","#2ecc71"]
sns.set_palette(sns.color_palette(my_palette))
ax = sns.countplot(x="Pgp_substrate", data=pgp_subs, order = ["Substrate", "Non-substrate"])
ax.set(ylabel='Number of compounds', xlabel=None, title ='Substrate to P-glycoprotein')
sns.despine(right=True, bottom = True)
In [9]:
#Pgp_active
replace_values_pgp_act = {0 : 'Inactive', 1 : 'Active'}
pgp_act = data.replace({"Pgp_active": replace_values_pgp_act})
my_palette = ["#e74c3c", "#2ecc71"]
sns.set_palette(sns.color_palette(my_palette))
ax = sns.countplot(x="Pgp_active", data=pgp_act, order = ["Active", "Inactive"])
ax.set(ylabel='Number of compounds', xlabel=None, title ='Activity againts P-glycoprotein')
sns.despine(right=True, bottom = True)
In [10]:
#BDNF, Reduction
replace_values_bdnf = {0 : 'Negative', 1 : 'Positive'}
bdnf = data.replace({"BDNF, Reduction": replace_values_bdnf})
my_palette = ["#e74c3c", "#2ecc71"]
sns.set_palette(sns.color_palette(my_palette))
ax = sns.countplot(x="BDNF, Reduction", data=bdnf)
ax.set(ylabel='Number of compounds', xlabel=None, title ='Reduction of BDNF')
sns.despine(right=True, bottom = True)
In [11]:
#Activity_Syn
replace_values_syn = {0 : 'Inactive/Unknown', 1: 'Active'}
syn = data.replace({"Activity_Syn": replace_values_syn})
my_palette = ["#e74c3c", "#2ecc71"]
sns.set_palette(sns.color_palette(my_palette))
ax = sns.countplot(x="Activity_Syn", data=syn, order=[ "Active", "Inactive/Unknown"] )
ax.set(ylabel='Number of compounds', xlabel=None, title ='Synaptogenesis')
sns.despine(right=True, bottom = True)
In [12]:
#Activity_NNF
replace_values_nnf = {0 : 'Inactive/Unknown', 1 : 'Potent'}
nnf = data.replace({"Activity_NNF": replace_values_nnf})
my_palette = ["#e74c3c", "#2ecc71"]
sns.set_palette(sns.color_palette(my_palette))
ax = sns.countplot(x="Activity_NNF", data=nnf, order = ["Potent", "Inactive/Unknown"])
ax.set(ylabel='Number of compounds', xlabel=None, title ='Neural Network Formation')
sns.despine(right=True, bottom = True)

Distribution of continuous variables

In [13]:
#SLogP
ax = sns.kdeplot(data['SLogP'], legend = False)
ax.set(xlabel = None, ylabel = None, title ='SLogP')
sns.despine(right=True, bottom = False)
In [14]:
#Cbrain/Cblood
ax = sns.kdeplot(data['Cbrain/Cblood'], legend=False)
ax.set(xlabel = None, ylabel = None, title ='Cbrain/Cblood')
sns.despine(right=True)
In [15]:
#Syn_EC30
ax = sns.kdeplot(data['Syn_EC30'], legend=False)
ax.set(ylabel = None, xlabel = 'EC30 (μM)', title ='Synaptogenesis')
sns.despine(right=True)
In [16]:
#Viability_EC30
ax = sns.kdeplot(data['Viability_EC30'], legend=False)
ax.set(ylabel = None, xlabel = 'EC30 (μM)', title ='Viability ATP Synaptogenesis')
sns.despine(right=True)
In [17]:
#NNF EC50min
ax = sns.kdeplot(data['NNF EC50min'], legend=False)
ax.set(ylabel = None, xlabel = 'EC50 min (μM)', title ='Neural Network Formation')
sns.despine(right=True)
In [18]:
#NNF EC50max
ax = sns.kdeplot(data['NNF EC50max'], legend=False)
ax.set(ylabel = None, xlabel = 'EC50 max (μM)', title ='Neural Network Formation')
sns.despine(right=True)
In [19]:
#Viability_LDH
ax = sns.kdeplot(data['Viability_LDH'], legend=False)
ax.set(ylabel = None, xlabel = 'μM', title ='Viability LDH Neural Network Formation')
sns.despine(right=True)

Correlations

In [20]:
# DNT vs SLogP
replace_values_dnt_slogp = {0 : 'Negative', 1 : 'Positive'}
dnt_slogp = data.replace({"DNT": replace_values_dnt_slogp})
my_palette = ["#e74c3c", "#2ecc71"]
sns.set_palette(sns.color_palette(my_palette))
ax = sns.swarmplot(x=dnt_slogp["DNT"], y=dnt_slogp["SLogP"], data=dnt_slogp)
ax.set(ylabel='SLogP', xlabel=None, title ='DNT vs Lipophilicity')
sns.despine(right=True, bottom = False)
In [21]:
ax = sns.violinplot(x=dnt_slogp["DNT"], y=dnt_slogp["SLogP"], data=dnt_slogp, inner=None)
ax = sns.swarmplot(x=dnt_slogp["DNT"], y=dnt_slogp["SLogP"], data=dnt_slogp,
                   color="white", edgecolor="gray")
ax.set(ylabel='SLogP', xlabel='DNT Classification', title ='DNT vs Lipophilicity')
sns.despine(right=True, bottom = False)
In [22]:
# DNT vs Cbrain/Cblood
replace_values_dnt_bbb = {0 : 'Negative', 1 : 'Positive'}
dnt_bbb = data.replace({"DNT": replace_values_dnt_bbb})
my_palette = ["#e74c3c", "#2ecc71"]
sns.set_palette(sns.color_palette(my_palette))
ax = sns.swarmplot(x=dnt_bbb["DNT"], y=dnt_bbb["Cbrain/Cblood"], data=dnt_bbb)
ax.set(ylabel='Cbrain/Cblood', xlabel=None, title ='DNT vs BBB')
sns.despine(right=True, bottom = False)
In [23]:
ax = sns.violinplot(x=dnt_bbb["DNT"], y=dnt_bbb["Cbrain/Cblood"], data=dnt_bbb, inner=None)
ax = sns.swarmplot(x=dnt_bbb["DNT"], y=dnt_bbb["Cbrain/Cblood"], data=dnt_bbb,
                   color="white", edgecolor="gray")
ax.set(ylabel='Cbrain/Cblood', xlabel='DNT Classification', title ='DNT vs BBB')
sns.despine(right=True, bottom = False)

Missing values

In [24]:
#Table with percentage
data_new = data.drop(columns=['Chemical', 'CASRN'])
missing = (data_new.isnull().sum(0)/97)*100
df = round(missing)
df
Out[24]:
ID                  0.0
DNT                 0.0
SLogP               2.0
BBB                 4.0
Cbrain/Cblood      10.0
Pgp_inhibition     34.0
Pgp_substrate      34.0
Pgp_active         34.0
BDNF, Reduction    62.0
Activity_Syn        0.0
Syn_EC30           31.0
Viability_EC30     31.0
Activity_NNF        0.0
NNF EC50min        11.0
NNF EC50max        11.0
Viability_LDH      19.0
dtype: float64
In [25]:
msno.bar(data)
Out[25]:
<matplotlib.axes._subplots.AxesSubplot at 0x243c0e2f370>
In [26]:
#Correlation heatmap 
#It measures nullity correlation
#How strongly the presence or absence of one variable affects the presence of another
msno.heatmap(data)
Out[26]:
<matplotlib.axes._subplots.AxesSubplot at 0x243bfad85e0>

Pairwise relationships

In [29]:
sns.pairplot(data)
Out[29]:
<seaborn.axisgrid.PairGrid at 0x1f6f5a85b20>

Correlation matrix

In [27]:
#Standardise continuous values
data_prep = data[['SLogP', 'Cbrain/Cblood','Syn_EC30', 
                  'Viability_EC30', 'NNF EC50min', 'NNF EC50max', 'Viability_LDH']]
sc = StandardScaler()
data_scaled = sc.fit_transform(data_prep)
data_scaled = pd.DataFrame(data_scaled)
data_scaled.columns = ['SLogP', 'Cbrain/Cblood','Syn_EC30', 
                       'Viability_EC30', 'NNF EC50min', 'NNF EC50max', 'Viability_LDH']

#Add other discrete values and details
discrete = data [['ID', 'Chemical', 'CASRN', 'DNT', 'BBB', 'Pgp_inhibition', 'Pgp_substrate', 'Pgp_active', 
  'BDNF, Reduction', 'Activity_Syn', 'Activity_NNF']]
data = discrete.join(data_scaled, lsuffix="_left", rsuffix="_right")
data = data[['ID', 'Chemical', 'CASRN', 'DNT', 'SLogP', 'Cbrain/Cblood', 'BBB', 'Pgp_inhibition', 'Pgp_substrate', 'Pgp_active', 'BDNF, Reduction', 'Activity_Syn', 'Syn_EC30', 
                       'Viability_EC30', 'Activity_NNF', 'NNF EC50min', 'NNF EC50max', 'Viability_LDH']]
In [28]:
#Standardised 
#This one used for manuscript
pearsoncorr = data.corr(method='pearson')
sns.heatmap(pearsoncorr, 
            xticklabels=pearsoncorr.columns,
            yticklabels=pearsoncorr.columns,
            cmap='RdBu_r',
            annot=False,
            linewidth=0.5)
Out[28]:
<matplotlib.axes._subplots.AxesSubplot at 0x243bfd2f220>
In [29]:
kendall = data.corr(method='kendall')
sns.heatmap(kendall, 
            xticklabels=pearsoncorr.columns,
            yticklabels=pearsoncorr.columns,
            cmap='RdBu_r',
            annot=False,
            linewidth=0.5)
Out[29]:
<matplotlib.axes._subplots.AxesSubplot at 0x243bfc08c10>
In [30]:
spearman = data.corr(method='spearman')
sns.heatmap(spearman, 
            xticklabels=pearsoncorr.columns,
            yticklabels=pearsoncorr.columns,
            cmap='RdBu_r',
            annot=False,
            linewidth=0.5)
Out[30]:
<matplotlib.axes._subplots.AxesSubplot at 0x243c101c6d0>

Parallel categories diagram

In [31]:
df = data[['DNT', 'BDNF, Reduction', 'Activity_Syn', 'Activity_NNF']]
fig = px.parallel_categories(df, color_continuous_scale=px.colors.diverging.Tealrose, color_continuous_midpoint=0)

fig.show()
In [32]:
%load_ext watermark
%watermark -n -u -v -iv -w

missingno 0.4.2
pandas    1.0.5
numpy     1.18.5
seaborn   0.10.1
last updated: Wed Aug 19 2020 

CPython 3.8.3
IPython 7.16.1
watermark 2.0.2
In [ ]: